GPU Performance limit thermal.

[blu3dragon]

Good to see hotspot added in the latest beta now.

I wonder if in this case you are hitting a vrm limit?
Will be interesting to see if new thermal pads fixes it.

 

[Badgerslayer7]

It will be. Still waiting for ek to send out the pads. It’s taking ages one email a day from them.

 

[Badgerslayer7]

So I’ve changed the thermal pads and I’m still getting thermal throttle yes. So who knows. Must be an error.

 

[bearpulla]

The same here with my EVGA 3090 XC3, GPU temp is 67, memory junction is @84, hitting thermal only

 

[SublimeXYZ]

This is very likely a driver bug or something. I have the MSI Gaming X Trio 3090 and I have managed to flip both power and thermal limits to yes by simply alt-tabbing in and out of Guild Wars 2 and believe me, Guild Wars 2 is DEFINITELY not pushing this card anywhere near thermals or power limits. My card's GPU core doesn't even go above 54°C while playing it, memory junction hovers around 62°C ~ 64°C and hot spot temp around 65°C give or take. Power is usually around 200W ~ 220W with a limit of 380W. I also read in other places that games which use Vulkan or DX12 can also cause this to happen when you alt-tab out of them, but Guild Wars 2 is DX9 so it doesn't seem like it is limited to just those two graphics APIs.

EDIT: Adding image. This time I even got max operating voltage to flip too, even though my GPU is slightly undervolted. Yes, with Guild Wars 2 and alt-tabbing in and out of it.

 

[edhunter]

I also see "fake/impossible" readings for thermal limit.
MSI 3080 Suprim X water cooled, latest nvidia drivers 466.11.
MSI Afterburner temp limit is also monitored, but its graph shows flat 0 all the time.

 

[Martin]

I will need to have a closer look at this. Please run the following build in Debug Mode and when you notice the Thermal limit being hit, close HWiNFO and upload the DBG file it created: www.hwinfo.com/beta/hwi64_703_4454.zip

 

[edhunter]

Okay will do that. It make take days before thermal flip happen. I will set alarm for it in order to see it.

 

[Martin]

Oh, that's too long, the DBG file will grow extremely large. It would be much better if this could be more quickly triggered.
When it takes so long to hit that limit, I'm wondering how have you been able to check in MSI Afterburner that there wasn't a short spike in the past? Have you logged all values?

 

[edhunter]

No I have not logged.
Yesterday I had in happen it less than hour. Yesterday evening I rebooted my pc for installing something, and some minutes later I played ACC and after quit I noticed that thermal has been triggered in hwinfo and then I just opened the AB monitor and extended the window maximal width. My monitor is large and it fits about one hour of data
Here is sample screenshot from today taken few minutes ago

 

[edhunter]

I will try to force it with debug enabled. Will experiment .. play a game do some alt-tab, run the quickminer, switch MSI AB profiles.
Is it a good idea to enable sensors logging in hwinfo together with debug? My guess is that thermal limit could be flipped only for a moment and is possible not to not be logged in the csv.

 

[Martin]

There's no need to enable CSV logging in HWiNFO, that won't provide me more details. I need to see the raw data in DBG when it happens, the only issue is that long runs will produce a very large file.

 

[edhunter]

Okay .. something strange is happening. Yesterday I was able to see temperature limit in MSI AB, but not in HWinfo. Today I got it in both. I don't believe that it really is thermal limit.
However I have debug file from today. It is 344MB uncompressed. I zipped it and uploaded it here - https://mega.nz/file/Io5FWS7D#65Y1KYDH3bekKP7htZIbC5fRctBgQUKr-u3ncH3P3Gg

 

[Martin]

The difference in capturing of the thermal limit is most likely because it's a very short spike and it depends on the polling interval in a particular tool.
The fact that MSI AB sees this too confirms that it's correctly reported. Why it happens - only NVIDIA knows exactly as this is what the drivers say. Might be a bug there...